Monitoring Methods for Moisture Content Evolution of Hydrophobic Modified Soil Subgrade During Service

Abstract

This invention belongs to the field of data analysis and processing technology, specifically relating to a method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life. The method includes: Step 1: Installing and numbering resistive soil moisture probes longitudinally, laterally, and at depth within the EL-NSM hydrophobic modified soil subgrade, establishing a node index table, a geometric adjacency table, and a voxel grid; Step 2: Performing directional energization in four directions according to an excitation grouping table to generate four types of acquisition windows; Step 3: Pairing droplet entries with pickling entries within the same acquisition window; forming a continuous body and a transition body, ultimately obtaining a voxel-level moisture content map; Step 4: Summarizing the voxel-level moisture content map into a monitoring zone moisture content map and a continuous chain segment list to continuously output the moisture content evolution results during the service life. This invention not only significantly improves the accuracy and reliability of moisture content monitoring but also provides dynamic predictions in both time and space dimensions.

Description

Technical Field

[0001] This invention belongs to the field of data analysis and processing technology, and particularly relates to the field of intelligent sensing system technology, specifically a method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life. Background Technology

[0002] Currently, in the field of civil engineering, there are numerous publicly available technologies for monitoring soil moisture content and its evolution in order to monitor the long-term service performance of highway and railway subgrades. Common methods mainly focus on resistive moisture sensing, time-domain reflectometry, frequency-domain reflectometry, and thermal impedance spectroscopy. These methods share the common feature of indirectly inferring the soil's moisture content through the sensor's response curve. For example, resistive sensing technology reflects moisture content by monitoring the difference in resistance between electrodes as moisture content changes. This method is simple in structure and low in cost, and has been widely used in research on farmland irrigation, slope stability, and subgrade moisture control. Time-domain reflectometry and frequency-domain reflectometry invert moisture content through the propagation characteristics of high-frequency electromagnetic signals in the soil, offering high accuracy. However, their devices are complex and costly, and they present difficulties in deployment and maintenance for long-term monitoring of large-scale infrastructure.

[0003] However, in the specific scenario of hydrophobic modified soil subgrades, existing technologies face several insurmountable problems. First, the introduction of hydrophobic modifying materials alters the capillary water migration and interfacial adsorption characteristics of the soil, making traditional methods based on linear resistance or capacitance changes inaccurate in depicting the true moisture distribution. This means that even with existing sensors, monitoring results may be biased due to the heterogeneity and nonlinear adsorption behavior of the modified soil. Second, most existing methods can only provide moisture content data for single points or local areas, lacking the ability to convert these discrete measurement results into a spatially continuous distribution. This is insufficient for the overall safety assessment of subgrades during their service life, as the infiltration and accumulation of moisture in the subgrade often exhibit significant spatial coupling characteristics, and relying solely on single-point data cannot reveal potential weakening paths. Summary of the Invention

[0004] The main objective of this invention is to provide a method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life. This method can not only significantly improve the accuracy and reliability of moisture content monitoring, but also provide dynamic prediction in both time and space dimensions.

[0005] To address the aforementioned technical problems, this invention provides a method for monitoring the moisture content evolution of hydrophobic modified soil subgrade during its service life. The method includes:

[0006] Step 1: Install and number resistive soil moisture probes in the longitudinal, transverse and depth directions within the EL-NSM hydrophobic modified soil subgrade, and establish a node index table, geometric adjacency table and voxel grid; collect the reference resistance of each node under dry conditions and generate a reference resistance map; generate an excitation grouping table based on the geometric adjacency table, which includes a main excitation group, a secondary excitation group and a guardrail group, and generate a list of neighboring nodes for each voxel;

[0007] Step 2: Perform directional energizing in four directions according to the excitation grouping table to generate four types of acquisition windows; align and linearly stretch the time series of each node using the reference resistance diagram, generating a directional sequence consisting of up, flat, and down curves, extracting and registering the three indices of the first offset, peak value, and platform start point; select the vertical nearest, horizontal nearest, and deep nearest nodes to form a three-bond group for each voxel based on the voxel's neighboring node list, and assemble the time stack under the four types of acquisition windows; generate candidate conduction path templates consistent with the acquisition window directions using the node index table and geometric adjacency table, and map the voxels to the template sequence according to the shared surface relationship;

[0008] Step 3: Pair droplet entries with pickling entries within the same acquisition window; within the same acquisition window, define the curve segments that show an initial increase followed by stabilization or an initial increase followed by a decrease relative to the main excitation polarity as pickling entries, and define the curve segments that show an initial decrease followed by stabilization or an initial decrease followed by an increase relative to the main excitation polarity as droplet entries, and pair them accordingly; based on the fingerprint category and the curve morphology of the three-bond group members, assign water-containing behavior labels to voxels on the candidate conduction paths and generate direction marks; perform riveting and penetration on voxels of the same type along the direction marks using the adjacent relationship of the shared surface to form a through body and a transition body, and finally obtain a voxel-level water content level map;

[0009] Step 4: Summarize the voxel-level moisture content maps into a monitoring zone moisture content map and a list of connected segments. Based on the spatial orientation of the connected segments, select the main incentive group, auxiliary incentive group, and guardrail group combination from the incentive grouping table for the next inspection, generate an incentive execution list, and repeat the execution in subsequent inspection cycles to continuously output the moisture content evolution results during the service period.

[0010] Further, step 1 specifically includes: deploying and numbering resistive soil moisture probes longitudinally, laterally, and in depth within the EL-NSM hydrophobic modified soil subgrade; establishing a node index table, which records the longitudinal, lateral, and depth coordinates of each numbered node; generating a geometric adjacency table based on the grid relationships of nodes in the three directions, listing the adjacent node numbers that share edges or faces with each node; dividing the monitoring volume into a voxel grid, and selecting one nearest node in each of the three directions for each voxel to form a list of adjacent voxel nodes; collecting the resistance of each node under dry conditions to form a reference resistance diagram; generating an excitation grouping table based on the row and column distribution of the node index table, which divides nodes into three categories: main excitation group, auxiliary excitation group, and guardrail group, and listing the energizing sequence in two sets of sequences, longitudinal and lateral.

[0011] Furthermore, in step 2, based on the excitation grouping table, two sets of window numbering tables, one for the longitudinal sequence and one for the transverse sequence, are generated within the monitoring zone; directional energizing excitation is sequentially performed in the four directions of longitudinal forward, transverse forward, longitudinal reverse, and transverse reverse to form four types of acquisition windows; for each acquisition window, the numbers of the main excitation group nodes, auxiliary excitation group nodes, and guardrail group nodes involved in the window are recorded according to the window numbering table; the resistance time series of the above nodes are acquired to form the original curve book.

[0012] Furthermore, in step 2, for each time series in the original curve book, the reference resistance of the corresponding node in the reference resistance diagram is found, and the reference resistance is subtracted from the time series point by point to obtain the repositioning curve. Within each acquisition window, the minimum value of the repositioning curve is taken as the zero point and the maximum value as one unit, and the repositioning curve is linearly stretched to obtain the amplitude stretching curve, which is then recorded in the window curve book. This step directly uses the reference resistance diagram to complete the repositioning reference, and the result is written into the window curve book.

[0013] Furthermore, in step 2, the process of generating the direction sequence includes: for each curve in the window curve book, generating a direction sequence point by point according to the size relationship of adjacent sampling points: the later sample is greater than the previous sample and is marked as up, the later sample is equal to the previous sample and is marked as flat, and the later sample is less than the previous sample and is marked as down; based on this, three position indicators are extracted and registered in the indicator table: the first offset sequence number is the position where the first consecutive up occurs at least twice, the peak sequence number is the turning point where the first up occurs and then the next down occurs, and the platform start sequence number is the position where the first consecutive flat occurs at least twice; based on the direction sequence and the three position indicators, the curve is determined to be one of the following four types and registered in the type table: up-flat type, characterized by at least three consecutive ups followed by at least two consecutive flats; up-down type, characterized by at least three consecutive ups followed by at least two consecutive downs; up-platform type, characterized by at least two consecutive ups, followed by at least two consecutive flats, followed by at least two more consecutive ups; up-up-platform type, characterized by at least four consecutive ups, followed by at least one flat.

[0014] Furthermore, in step 2, each voxel in the voxel mesh is traversed, and the list of nearest voxel nodes is called to extract the triple bond group of that voxel. The triple bond group consists of the vertical nearest node, the horizontal nearest node, and the deep nearest node. Under the four types of acquisition windows, the entries of the triple bond group members in the index table and the morphology table are read respectively, assembled into a time stack according to the window order, and written to the voxel workbook. This step directly uses the voxel mesh and the list of nearest voxel nodes to complete the binding of voxel and node features.

[0015] Furthermore, in step 2, based on the coordinate relationships in the node index table and the adjacency pairs in the geometric adjacency table, three types of candidate pathway templates are generated along the vertical, horizontal, and depth directions, respectively. Each template consists of a sequence of voxels representing a shared surface, and the voxel sequences inherit the adjacency relationships of the shared surface at the voxel level through the geometric adjacency table. For each acquisition window, a path template consistent with the direction of that window is selected, and the voxel sequences in the path template are registered as the candidate pathway set for that window. This step explicitly uses the node index table to generate the direction and uses the geometric adjacency table to implement the adjacency relationships at the voxel level.

[0016] Further, in step 2, within the voxel workbook, drop and draw segments for each voxel are paired within the same acquisition window to form response fingerprint pairs. Each drop and draw segment entry consists of an initial offset number, a peak number, and a morphological type. The response fingerprint pair category is determined according to the following rules and written into the fingerprint matrix: Forward type: the initial offset number of the drop segment is earlier than that of the draw segment; Backward type: the initial offset numbers of the drop and draw segments are the same, and the peak number of the drop segment is later than that of the draw segment; Balanced type: the initial offset numbers and peak numbers of the drop and draw segments are identical. Simultaneously, referring to the guardrail groups in the excitation grouping table, the amplitude range of the guardrail group curve and the amplitude range of the triple bond group curve are calculated within the same window. If the amplitude range of the guardrail group is less than half of the amplitude range of the triple bond group, then the voxel is registered with a guardrail pass mark for subsequent ontology response confirmation in label assignment. This step explicitly uses the contents of the voxel workbook, the excitation grouping table, and the window curve book.

[0017] Furthermore, in step 3, within each acquisition window, for voxels in the candidate conduction path set, their category in the fingerprint matrix and type in the morphology table are read. Combined with the guardrail passmark, water-containing behavior labels and direction marks are generated at the voxel level according to the following rules and written to the label layer: Axial penetration: When the window direction is vertical and the vertical member in the triple bond group is either forward-pushing or balanced, and the morphology type of the vertical member is type I or IV, while the first segment of the direction sequence of the horizontal and depth members is upward, then the voxel is marked as axial penetration, and the direction mark is vertical and consistent with the window direction; when the window direction is horizontal and the horizontal member in the triple bond group is either forward-pushing or balanced, and the morphology type of the horizontal member is type I or IV, while the first segment of the direction sequence of the vertical and depth members is upward, then the voxel is marked as axial penetration, and the direction mark is vertical and consistent with the window direction; when the window direction is horizontal and the horizontal member in the triple bond group is either forward-pushing or balanced, and the morphology type of the horizontal member is type I or IV, while the first segment of the direction sequence of the vertical and depth members is upward, then the voxel is marked as axial penetration, and the direction mark is vertical. The markings are horizontal and consistent with the window direction; oblique penetration: when the voxel is in a forward-pushing type under both the vertical and horizontal windows, and the order of the peak numbers in the two windows is inconsistent, the voxel is marked as oblique penetration, and the direction marking is a combination of vertical and horizontal; interface seepage: when the deep member of the triple bond group under the deep window is in a backflow type and the morphology type is type 2, the voxel is marked as interface seepage, and the direction marking is deep; bottom backflow: when the peak number order of the triple bond group under the deep window is led by the deep member, and the morphology type of both the vertical and horizontal members is type 1, the voxel is marked as bottom backflow, and the direction marking is deep; chamber water storage: when the response fingerprint pairs of the voxel are all balanced under all four types of windows, and at least two of the triple bond group members are type 1, the voxel is marked as chamber water storage, and the direction marking is omnidirectional.

[0018] Furthermore, in step 3, voxels marked as axial or oblique penetration in the label layer are used as seed voxels. Based on the geometric adjacency table, penetration chains are generated at the voxel level using shared face adjacency relationships, pointing along their respective direction marks. When adjacent voxels at the front end of a penetration chain segment have the same direction mark and are marked as axial or oblique penetration in the label layer, that voxel is incorporated into the chain segment. When three or more penetration chains meet on the same voxel and their direction marks are different pairwise, that voxel is registered as a node and a riveting operation is performed. The riveting operation binds the end of the merging chain segments to the node, forming a component of a single penetration body. Subsequently, voxels sharing a face with the penetration body and labeled as interface seepage or bottom backflow are incorporated into the transition body. The remaining voxels in contact with the penetration body or transition body through shared faces are subjected to joint filling to form a filling body. The through-body, transition body, and infill body are mapped to a voxel-level moisture content gradation map: the through-body corresponds to a high grade, the transition body to a medium grade, and the infill body to a low grade. This step explicitly uses a geometric adjacency list and voxel mesh to complete spatial penetration and solidification.

[0019] The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life, as proposed in this invention, has the following beneficial effects: it can output dynamic moisture distribution results stably and over a long period of time under complex environments, which has significant beneficial effects.

[0020] First, this method achieves a complete representation of the spatial structure by deploying resistive moisture probes inside the subgrade and constructing a node index table, a geometric adjacency table, and a voxel grid. This allows the monitored objects to no longer be limited to single-point measurements, but to establish a continuous distribution map at the voxel level, thereby improving spatial resolution and overall integrity.

[0021] Secondly, this invention controls the power-on sequence through an excitation grouping table, and performs node time series repositioning, stretching, and direction sequence generation under four types of acquisition windows, so that each response curve can be compared under a unified benchmark, significantly reducing the error caused by soil heterogeneity.

[0022] Furthermore, this invention assembles a time stack at the voxel level and maps it using candidate conduction path templates, thereby highly binding voxels and node features. On this basis, response fingerprint pairs are generated by pairing droplet entries and siphon entries, and the water-bearing behavior labels and direction marks are determined by combining curve morphology. This allows the invention to reveal the true infiltration paths and evolutionary characteristics of water in the longitudinal, lateral, and deep directions. This process effectively solves the limitation of existing technologies in describing dynamic evolution laws.

[0023] Finally, through the riveting and penetration strategy, this invention can connect similar voxels along the direction markings to form a through body, and combine it with the transition body and the filling body to form a stable voxel-level moisture content map. Then, by summarizing the data by region, a monitoring zone moisture level map and a list of through chain segments are generated, which enables the monitoring system to have the ability to adaptively adjust the excitation execution and continuously track the expansion and contraction of the infiltration channel.

[0024] Overall, this invention not only significantly improves the accuracy and reliability of moisture content monitoring, but also provides dynamic predictions in both time and space dimensions, providing technical support for the long-term service safety assessment and early warning of roadbeds. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0026] Figure 1 A schematic diagram of the method flow for monitoring the moisture content evolution of hydrophobic modified soil subgrade during service life, provided in an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of a typical response curve under four-directional excitation provided in an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram comparing the resistance response of the three-bond group nodes under different excitation directions, as provided in an embodiment of the present invention. Detailed Implementation

[0029] The method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0030] refer to Figure 1 A method for monitoring the evolution of moisture content in hydrophobically modified soil subgrade during its service life, the method comprising:

[0031] Step 1: Install and number resistive soil moisture probes in the longitudinal, transverse and depth directions within the EL-NSM hydrophobic modified soil subgrade, and establish a node index table, geometric adjacency table and voxel grid; collect the reference resistance of each node under dry conditions and generate a reference resistance map; generate an excitation grouping table based on the geometric adjacency table, which includes a main excitation group, a secondary excitation group and a guardrail group, and generate a list of neighboring nodes for each voxel;

[0032] During implementation, the monitoring zone boundaries were first determined on the EL-NSM hydrophobic modified soil subgrade. The longitudinal direction of the subgrade was defined as longitudinal, the direction perpendicular to the subgrade centerline was defined as transverse, and the direction pointing inwards from the subgrade was defined as depth. A rectangular coordinate system was established along the longitudinal, transverse, and depth directions, using the intersection of the subgrade centerline and the selected station as the origin, to uniformly record the location of the resistive soil moisture probes and the spatial range of the voxel grid. The direct benefit of using a unified coordinate system is that the subsequent geometric adjacency table and voxel grid can be seamlessly aligned, avoiding the accumulation of errors caused by repeated conversions.

[0033] Within the monitoring zone, boreholes are laid out according to a regular grid pattern in the longitudinal, transverse, and depth directions, and resistive soil moisture probes are installed at predetermined locations. Only one resistive soil moisture probe is installed at each location, with the probe's sensitive area completely penetrating the undisturbed soil layer. To reduce the dispersion of contact resistance, the borehole walls are backfilled with fine-grained material of the same origin as the undisturbed soil and compacted in layers, ensuring that the current path is primarily governed by the soil moisture content and not altered by artificial filler. A three-segment numbering method is used to number all probes. The first segment is the longitudinal sequence number, the second is the transverse sequence number, and the third is the depth sequence number, separated by hyphens. For example, a node numbered 12-05-03 represents a node with a longitudinal sequence number of 12, a transverse sequence number of 05, and a depth sequence number of 03. The three-segment numbering corresponds one-to-one with the three-dimensional coordinates, allowing for direct location during subsequent cross-references between the node index table, geometric adjacency table, and voxel grid, reducing ambiguity.

[0034] The longitudinal, lateral, and depth coordinates of each numbered node are recorded using measured coordinates to form a node index table. The node index table also records the node's burial depth, installation date, and zone number, used for subsequent filtering of local areas or comparison of different layers. The direct benefit of this unified recording is that any step requiring node location or aggregation of nodes in the same layer can be completed in one step using the node index table. Using the three-dimensional coordinates from the node index table as input, each node is searched for its adjacent nodes in the longitudinal, lateral, or depth directions. The criterion is that they are adjacent in only one direction and have the same sequence number in the other two directions. Two nodes that meet this criterion share an edge or face in space. All adjacent node numbers of each node are registered as the node's adjacency list, which is then compiled into a geometric adjacency list. The geometric adjacency list defines the possible dominant propagation relationships of the current. The direct benefit of using a shared edge or face determination method is that the adjacency relationship is consistent with the shared face relationship of the voxel mesh, which can be directly reused when generating candidate conduction paths and riveting connections.

[0035] Based on the three-dimensional dimensions and node distribution of the monitoring zone, voxel units are divided along the longitudinal, transverse, and depth directions to form a voxel mesh. The voxel side length is selected to match the node spacing, so that each voxel can be represented by several surrounding nodes. The direct benefit of matching the side length with the node spacing is that the voxel will not cross too many current path turning points, reducing spatial aliasing. A voxel number is generated for each voxel, and the voxel numbering rule maintains a three-segment consistency with the node numbering, facilitating mutual lookup and mapping between voxels and nodes.

[0036] Data collection was conducted during periods of continuous drought and dry roadbed surface to stabilize soil pore water content at a low level. The direct advantage of selecting a dry period for baseline data collection is that resistivity is primarily determined by the skeleton contact and a small amount of pore water; any subsequent increase in moisture content will result in a decrease in resistance, exhibiting a singular directional relationship, facilitating threshold determination and time sequence analysis. Each numbered node was sampled continuously, three times, with fixed short intervals between samples. If two of the three data points were close and the deviation was within a predetermined small range, the more frequently occurring value was taken as the baseline resistance for that node; if the three values ​​differed significantly, a fourth sample was added, and the baseline value was determined using the same rules. This method directly suppresses outliers caused by occasional pulses and contact jitter. The baseline resistances of all nodes were filled into a three-dimensional array according to node number and spatial location to obtain a baseline resistance map. The baseline resistance map provides a unified reference for subsequent repositioning and amplitude stretching, making the time series of different nodes comparable.

[0037] Based on the node index table and geometric adjacency table, nodes with the same vertical coordinate and depth coordinate are grouped into a column along the vertical direction, forming a vertical column group; nodes with the same horizontal coordinate and depth coordinate are grouped into a row along the horizontal direction, forming a horizontal row group. Within the vertical column group and horizontal row group, each column or row is selected as an entry in the main excitation group, and the columns or rows directly adjacent to that column or row are selected as corresponding entries in the auxiliary excitation group, forming a main-auxiliary pair. The main excitation group is responsible for generating the directional current field, while the auxiliary excitation group provides symmetrical constraints on both sides to reduce current diffusion in non-target directions. Nodes surrounding the outer edge of the monitoring zone and nodes near the boundary are grouped into the guardrail group. The guardrail group does not undertake the dominant excitation when energized; it only performs synchronous acquisition to identify boundary leakage current and external disturbances. When the amplitude of the guardrail group after benchmarking is significantly lower than that of the main excitation region, it can be confirmed that the main response of this excitation mainly occurs within the monitoring zone. A power-on sequence is set for each primary and secondary combination, proceeding sequentially from one end of the monitoring zone to the other. Independent sequence lists are generated for both the longitudinal and lateral directions, and then merged to form an excitation grouping table. The direct benefit of sequential progression is that the order of responses can be determined in subsequent steps, thereby inferring the propagation trend of moisture along the target direction.

[0038] For each voxel in the voxel mesh, the node closest to its geometric center is found in the longitudinal, lateral, and depth directions. One node is selected in each direction to form a triple bond group for that voxel, and these triple bond groups are registered as a list of neighboring voxels using a fixed field order. When two candidate nodes are equidistant, the node with the smaller number is selected to ensure the uniqueness of the neighboring voxel list during repeated generation. The direct benefit of the triple bond group being composed of the three nearest nodes is that it decouples the dominant response in each direction to the corresponding node, which helps distinguish between axial penetration, oblique penetration, and depth behavior, reducing directional confusion in subsequent steps. When there are irregular areas in the terrain or structure, the voxel boundary can be fine-tuned along the structural interface to make the voxel segmentation fit the real interface; the voxel numbering still uses the three-segment rule, and the generation steps of the node index table and geometric adjacency table remain unchanged. This method can improve the positioning accuracy near the interface while maintaining consistency in terminology and data structure. In sections with high sensor density, the list of nearest voxel nodes can be supplemented with records of the next nearest nodes for subsequent data filtering without changing the three-key group determination rules. If the data of the primary nearest node is missing, the data of the next nearest node can be temporarily referenced to supplement the data, so as to maintain the continuity of voxel-level data.

[0039] Step 2: Perform directional energizing in four directions according to the excitation grouping table to generate four types of acquisition windows; align and linearly stretch the time series of each node using the reference resistance diagram, generating a directional sequence consisting of up, flat, and down curves, extracting and registering the three indices of the first offset, peak value, and platform start point; select the vertical nearest, horizontal nearest, and deep nearest nodes to form a three-bond group for each voxel based on the voxel's neighboring node list, and assemble the time stack under the four types of acquisition windows; generate candidate conduction path templates consistent with the acquisition window directions using the node index table and geometric adjacency table, and map the voxels to the template sequence according to the shared surface relationship;

[0040] During implementation, based on the excitation grouping table, the main excitation group and auxiliary excitation group are selected sequentially for directional energization in the order of longitudinal forward, lateral forward, longitudinal reverse, and lateral reverse. The guardrail group is only sampled and not energized throughout the process. The reason for using four directions is that the current field has a directional offset in anisotropic media. The paired forward and reverse directions can offset boundary and contact errors, and the intersection of longitudinal and lateral directions can distinguish the axial and lateral propagation trends. Each directional energization is timed from the start of energization, and the resistance time series of all nodes involved in this excitation is recorded simultaneously, forming a type of acquisition window for that direction. The criterion for the end of the acquisition window is that the direction sequence is flat for 20 consecutive sampling points, or that after an initial upward movement followed by a downward movement, the upward movement does not reappear in the next 10 sampling points. This ending method can close the window in time after the signal dynamic process is completed, preserving both rising and falling information while avoiding prolonging the static segment of the stable phase. Four types of acquisition windows are generated for each of the four directions, and a unique window number is assigned to each energization in each type of acquisition window according to the energization sequence of the excitation grouping table. The window number corresponds one-to-one with the power-on sequence, which facilitates the retention of the sequence in the time stack and is used to determine the propagation leader direction.

[0041] For each node within each acquisition window, its reference resistance in the reference resistance diagram is read. The time series is then subtracted point by point from the reference resistance to obtain the repositioning curve. Repositioning processing eliminates absolute bias between different nodes, making relative changes under the same operating condition comparable across different nodes. Within the same acquisition window, each repositioning curve undergoes linear stretching: the minimum value of the curve within the window is taken as zero, and the maximum value is taken as one unit. All sampling points are mapped to the range of zero to one proportionally to this interval. Linear stretching unifies the amplitude of each node to the same scale, ensuring that subsequent direction sequences and the determination of the three sequence numbers are not affected by differences in the amplitude of individual nodes. The processed curves are registered in the window curve book according to the node and window number.

[0042] Within the window curve book, a direction sequence is generated for each curve according to the size relationship of adjacent sampling points: a later sample greater than the previous sample is marked as "up," a later sample equal to the previous sample is marked as "flat," and a later sample less than the previous sample is marked as "down." The direction sequence is stored in the order of sampling points to ensure a one-to-one correspondence with the original data. Three indices are extracted and registered in the index table. The first offset indices are the positions where the curve first appears consecutively "up," with at least two consecutive "up" occurrences; the peak indices are the positions where the curve first appears "up" and then "down"; and the plateau start indices are the starting positions where the curve first appears "flat," with at least two consecutive "flat" occurrences. The advantage of using continuous determination instead of single-point determination is that it can suppress mislabeling caused by single disturbances, ensuring that the three indices stably represent the starting point of the rise, the first peak, and the beginning of the stable region.

[0043] For each voxel in the voxel mesh, the list of nearest neighbors is read, and the vertical nearest node, horizontal nearest node, and deep nearest node are selected to form a three-bond group. The three-bond groups are registered in a fixed field order to ensure no overlap occurs during subsequent reads. Under four types of acquisition windows, the three serial numbers of each node in the three-bond group in the index table and the direction sequence summary in the window curve book are read sequentially. The window identifier, node number, initial offset serial number, peak serial number, platform start serial number, and direction sequence summary are written into the time stack in a fixed field order. The time stack is arranged according to the window number order. This assembly can completely preserve the temporal sequence relationship of the four directions, facilitating subsequent comparison of response differences in different directions on candidate conduction paths.

[0044] The coordinates in the node index table are read and combined with the adjacency pairs in the geometric adjacency table to generate candidate conduction path templates along the vertical, horizontal, and depth directions. Each template consists of a series of voxels with shared surfaces arranged sequentially, and the shared surface relationships between adjacent voxels are recorded. Using shared surface relationships instead of shared point relationships ensures that voxels on the path form a physically continuous conduction channel, avoiding virtual connections caused by single-point touches. For each type of acquisition window, candidate conduction path templates with the same direction as that window are selected as the template set for that type of window. For example, in the vertical forward acquisition window, the vertical template set is selected; in the horizontal reverse acquisition window, the horizontal template set is selected. The consistent direction selection ensures that the current field dominated by the excitation is in the same direction as the template, thereby improving the discriminative power when identifying conduction voxels on the template.

[0045] Under each type of acquisition window, voxels in the voxel grid are mapped to the candidate conduction path template sequence of that type of window according to their geometric positions. The mapping rule is as follows: when a voxel coincides with or is adjacent to a voxel in the path under a shared surface relationship, and the center line connecting the two is basically consistent with the template orientation, the voxel is assigned to the voxel sequence of that path. When a voxel simultaneously satisfies the mapping conditions of two parallel paths, it is preferentially assigned to the path whose center line is closer to the voxel. This classification method allows the path sequence to be closer to the actual main current channel, reducing mutual interference between paths. After mapping, a set of candidate conduction paths consistent with the direction of the acquisition window is formed, providing an ordered spatial carrier for water-bearing behavior label assignment and connection. In areas with complex surface geometry or strong boundary disturbances, the termination criterion of the acquisition window can be switched from the stationary criterion of the direction sequence to the relative criterion of amplitude variation. For example, the window ends when the maximum span of the amplitude within 20 consecutive sampling points does not exceed a small percentage of the maximum amplitude within the window. Using a relative criterion can avoid excessively long windows in low amplitude responses. Before linear stretching, a three-point median smoothing can be performed on the repositioning curve to remove isolated spikes, and then stretching can be performed based on the minimum and maximum values. Smoothing before stretching can reduce the amplification effect of a single outlier on the scale mapping and improve the stability of the direction sequence and the three indices.

[0046] Step 3: Pair droplet entries with pickling entries within the same acquisition window; within the same acquisition window, define the curve segments that show an initial increase followed by stabilization or an initial increase followed by a decrease relative to the main excitation polarity as pickling entries, and define the curve segments that show an initial decrease followed by stabilization or an initial decrease followed by an increase relative to the main excitation polarity as droplet entries, and pair them accordingly; based on the fingerprint category and the curve morphology of the three-bond group members, assign water-containing behavior labels to voxels on the candidate conduction paths and generate direction marks; perform riveting and penetration on voxels of the same type along the direction marks using the adjacent relationship of the shared surface to form a through body and a transition body, and finally obtain a voxel-level water content level map;

[0047] During implementation, based on the window curve book and index table, segment identification is performed on the linearly stretched repositioning curve of each node within the same acquisition window. Using the polarity of the main excitation as a reference, continuous upward movements in the direction sequence are considered increasing, continuous downward movements are considered decreasing, and continuous flat movements are considered stable. When a continuous segment of the curve within the acquisition window shows an initial increase followed by stabilization or an initial increase followed by a decrease, it is registered as a "drip segment" entry; when a continuous segment of the curve within the acquisition window shows an initial decrease followed by stabilization or an initial decrease followed by an increase, it is registered as a "drop segment" entry. An initial increase or decrease must have at least two consecutive sampling points, followed by stabilization must have at least two consecutive sampling points, and followed by an increase or decrease must have at least two consecutive sampling points. Using continuous points as the criterion can suppress misjudgments triggered by single-point noise, ensuring that the entries correspond to the actual moisture migration trend within the conductive path. For each entry, the window identifier, node number, initial offset sequence number, peak sequence number, and platform start-up sequence number are recorded, and the curve shape type is appended to the entry. The initial offset number and peak number are used to express the order of initiation and transition, and the plateau start number is used to express the time when the stable interval appears. These three together define the time position of the entries, which facilitates subsequent pairing and sorting at the voxel level.

[0048] The process of pairing entries and generating response fingerprints within the same acquisition window includes: for each voxel in the voxel workbook, reading the droplet and drawlet entries of the three-key group members within the same acquisition window. If a member only generates a single type of entry, the missing side is marked as having no entry and deferred to the optional implementation method for processing. Within the same acquisition window, using the first offset sequence number as the primary key, droplet and drawlet entries of the same node are paired one by one to form response fingerprint pairs. If there are multiple droplet entries or multiple drawlet entries of the same node, the pair with the smallest first offset sequence number is selected for pairing first, and entries that are not paired do not participate in subsequent determinations within this window. Prioritizing the matching of the earliest started pair can capture the primary response of water movement driven by the current field and reduce the interference of secondary ripples. For paired response fingerprint pairs, the fingerprint category is determined according to the following rules: when the first offset number of the drop segment entry is earlier than the first offset number of the draw segment entry, it is registered as a forward-pushing type; when the first offset numbers of both are the same and the peak number of the drop segment entry is later than the peak number of the draw segment entry, it is registered as a back-drawing type; when both the first offset number and the peak number are the same, it is registered as a balanced type. The fingerprint category expresses the difference in water moving towards the main current channel, flowing back to the interface, or stagnating locally through relative order and contrast.

[0049] Voxel sequences are processed path-by-path in the candidate conduction path set. For each voxel, the fingerprint category and curve morphology type of the three-bond group members under the four acquisition windows in the voxel workbook are read, and the guardrail pass mark is read for body response confirmation. In the vertical window, when the vertical member of the three-bond group is of the forward-pushing or balanced type, and the curve morphology type of the vertical member is the upper flat type or upper-upper plateau type, and the first segment of the direction sequence of the lateral and depth members is upward, then the voxel is assigned the water-bearing behavior label as axial penetration, and the direction mark is set to vertical, with the direction consistent with the current direction of the window. This assignment expresses the state of continuous conduction of water in the main current direction, and the direction mark is used for the advancement direction of subsequent riveting penetration. In the horizontal window, when the lateral member of the three-bond group satisfies the condition of symmetry with the previous one, the voxel is assigned the water-bearing behavior label as axial penetration, and the direction mark is set to horizontal. Through the symmetry determination of vertical and horizontal, the conducting voxels formed by lateral seepage can be identified. When the voxel is of the forward-pushing type in both the longitudinal and transverse windows, and the peak sequence numbers in the two windows are inconsistent, the water-bearing behavior is labeled as oblique penetration, and the direction symbol is set to a combination of longitudinal and transverse. The inconsistent peak sequence indicates that water connects two main directions obliquely, and the combined direction symbol provides a basis for subsequent alternating advancement in the two directions. When the deep member of the triple bond group in the depth window is of the backflow type and the curve morphology is up-down, the water-bearing behavior is labeled as interface infiltration, and the direction symbol is set to depth. This situation indicates that there is an upward backflow process at the deep interface, and it is necessary to prioritize expansion in the depth direction during subsequent penetration to cover the voxel affected by the interface. When the peak sequence number of the triple bond group in the depth window is led by the deep member, and the curve morphology of both the longitudinal and transverse members is flat upward, the water-bearing behavior is labeled as bottom backflow, and the direction symbol is set to depth. This situation corresponds to the state where water in the lower part returns upward and tends to plateau in the upper part. When the response fingerprint pairs of the voxel are all balanced under the four acquisition windows, and at least two of the curve morphology types of the three-bond group members are flat, the water-containing behavior is labeled as chamber water storage, and the direction symbol is set to omnidirectional. This indicates that a local water storage cavity is formed around the voxel, and subsequent penetration is mainly achieved by the surrounding conductive voxels.

[0050] Using voxels in the label layer with water-containing behavior labels of axial penetration and oblique penetration as seed voxels, and employing shared-face adjacency relationships at the voxel level according to the geometric adjacency list, penetration chains are generated by advancing along their respective direction marks. Shared-face advancement ensures that the current path between chains has a continuous cross section, reducing virtual connections caused by point contacts. When adjacent voxels at the front end of a penetration chain have the same direction mark and their water-containing behavior label is axial penetration or oblique penetration, the voxel is incorporated into the penetration chain. If the direction marks of adjacent voxels are a combination of longitudinal and transverse, alternating longitudinal and transverse incorporation is performed during advancement, causing the chain to extend along an oblique broken line. Alternating incorporation can approximate the actual oblique guiding path without introducing oblique voxel templates. When three or more penetration chains meet at the same voxel and their direction marks are different from each other, the voxel is registered as a node and a riveting operation is performed. The riveting operation establishes a fixed connection between the end of the merging chain and the node, making multiple chains part of a single penetrating body. The placement of nodes stabilizes connectivity at complex junctions, preventing chain breaks during updates. After the through-body is formed, voxels sharing the through-body's surface with water-bearing behavior tagged as interfacial seepage or bottom backflow are incorporated into the transition body. The transition body represents secondary water activity regions tightly coupled to the dominant channel. For the remaining voxels in contact with the through-body or transition body via shared surfaces, a filler body is formed to improve spatial continuity of voxel-level water content. Filling occurs only when there are three or more adjacent contacts, avoiding overexpansion of wide-area, low-coupling regions.

[0051] The continuous body, transition body, and filling body are mapped to a voxel-level water content grading map, where the continuous body corresponds to a high grade, the transition body to a medium grade, and the filling body to a low grade. This grading method directly reflects the differences in continuous conductivity and coupling strength, facilitating a one-to-one correspondence with the water content grading map of the monitoring zone. The voxel-level water content grading map and the list of continuous chain segments are output together to complete the summary of the water content grading map of the monitoring zone and the derivation of the incentive execution list for the next inspection. When there are multiple pairable droplet and pickling segment entries within the same acquisition window, after the pair with the smallest initial offset number is paired, one pair can be added sequentially in ascending order of the initial offset number to refine the description of the multi-peak response within the voxel. If the added pairing results in inconsistent fingerprint categories, the initial pairing result is used as the decisive category under that window, and the added result is registered as auxiliary information and does not participate in the final assignment of the water content behavior label. When a member of a triple bond group has no entry in a certain acquisition window, the entry for that window can be supplemented from the next nearest node in the same direction from the voxel's neighboring node list, and marked as a replacement entry in the voxel workbook. When entries for at least two members of the triple bond group are missing in the same window, that voxel will not participate in the assignment in that window, but will only participate in other windows. When the boundary region's guardrail is not satisfied by the marker in multiple adjacent voxels, the water content behavior labels for these voxels can be assigned later. First, the insertion and connection of the inner domain voxels can be completed, and then the boundary voxels can be gradually determined according to the outward trend of the connected body, so as to reduce the impact of external interference on the label determination.

[0052] Step 4: Summarize the voxel-level moisture content maps into a monitoring zone moisture content map and a list of connected segments. Based on the spatial orientation of the connected segments, select the main incentive group, auxiliary incentive group, and guardrail group combination from the incentive grouping table for the next inspection, generate an incentive execution list, and repeat the execution in subsequent inspection cycles to continuously output the moisture content evolution results during the service period.

[0053] During implementation, the voxel-level moisture content map and the list of connected segments are read, and voxels and segments are assigned to their respective monitoring zones according to the monitoring zone boundaries. Voxel assignment is based on the monitoring zone where the voxel's geometric center is located; for adjacent zones, assignment is based on proximity to the geometric center to eliminate ambiguity. For each monitoring zone, the number and proportion of voxels at high, medium, and low levels in the voxel-level moisture content map are statistically analyzed, and connected segments crossing the zone and their directional symbols are identified. The direct benefit of this approach is that it juxtaposes spatially strongly connected connectivity information with area statistical information, preserving both directional connectivity and overall comparative capabilities at the zone scale.

[0054] For each monitoring zone, the dominant level is first determined. If the proportion of high-level voxels reaches or exceeds 30% and forms a continuous band structure with a length of no less than 3 voxels within the zone, the dominant level of the zone is set to high. If the above conditions are not met but the proportion of medium-level voxels reaches or exceeds 40%, the dominant level is set to medium. In other cases, the dominant level is set to low. The order of looking at the proportion first and then the continuity can avoid a single point of high response raising the entire zone, while ensuring that the risk of band formation is marked in time. The dominant level is then processed for connectivity consistency. Isolated high-level patches with an area of ​​less than 9 voxels are downgraded to medium level to remove isolated noise. For zones traversed by a continuous chain segment and with a dominant level of medium, the voxel band traversed by the continuous chain segment is retained as a high-level band in the monitoring zone moisture level map to reflect the true conductive skeleton. The direct benefit of this processing is that it avoids obscuring key channels while maintaining the simplicity of the overall classification. The monitoring zone moisture level map is output. Each monitoring zone is colored with a single dominant level, and the high-level band traversed by the continuous chain segment is superimposed. The output uses a diagram with the same partition number and direction symbol to facilitate a one-to-one correspondence with the entries in the excitation grouping table.

[0055] The entire structure is divided into several connected segments, using nodes as dividing points. Each connected segment starts at a node or boundary voxel and ends at the next node or boundary voxel. This division method ensures stable directional markings and a single geometric orientation within each segment, facilitating matching with the main and auxiliary excitation groups. A list of entries is created for each connected segment, with a fixed field order including monitoring partition number, starting voxel number, ending voxel number, directional marking, segment length, number of voxels traversed, adjacent monitoring partition numbers, and the number of acquisition window categories involved. Segment length is represented by a voxel sequence count, ensuring reversible correspondence with voxel-level data. When a connected segment crosses a monitoring partition boundary, an entry is registered in the connected segment lists of both monitoring partitions, referencing each other through the adjacent monitoring partition number field. The direct benefit of dual-sided registration is that cross-table searches are unnecessary when performing excitation orchestration within a partition, and the corresponding segment on the opposite side can be quickly located when coordinated inspection is required.

[0056] For each monitoring zone, the moisture level map and the list of continuous chain segments for that zone are read. Combined with the excitation grouping table, the main excitation group, auxiliary excitation group, and guardrail group combination for the next inspection are determined. The selection of the main excitation group prioritizes directional consistency. For continuous chain segments with a vertical direction, the main excitation group entry with the most overlapping columns is selected first. When multiple entries have the same number of overlapping columns, the entry in the column where the geometric center of the continuous chain segment is located is selected. This selection ensures that the current principal axis is aligned with the existing conduction direction, thus most sensitively capturing changes at the leading edge of the continuous body in the next inspection. The selection of the auxiliary excitation group prioritizes symmetry constraints. For a determined main excitation group entry, the entries adjacent to it on both sides are selected as the auxiliary excitation group; if one side is the boundary of the monitoring zone or has missing nodes, the nearest available entry on the inner side is used instead. Symmetry constraints can suppress current divergence in non-target directions, improving the resolution of directional identification. The selection of the guardrail group prioritizes boundary protection. Outside the coverage areas of the main and auxiliary excitation groups, one or two rings of guardrail entries are selected for synchronous acquisition, prioritizing coverage of the area where the extension direction of the through-chain segment intersects with the boundary of the monitoring zone. This configuration enables timely detection of leakage and external disturbances at high gradient boundaries, reducing the detection of false connections. The excitation sequence follows a strategy combining risk-first and spatial adjacency. First, monitoring zones with higher dominant levels and their through-chain segments are placed at the top of the execution list; within the same level, entries with longer chain lengths and cross-zone references with adjacent monitoring zones are executed first; then, the remaining entries are arranged according to spatial adjacency, ensuring continuous advancement of the power-on path and shortening the stabilization time caused by power-on switching. An excitation execution list is generated. The list records the sequence number, monitoring zone number, acquisition window direction, main excitation group entries, auxiliary excitation group entries, guardrail group entries, target through-chain segment number, and target voxel range description in a fixed field order. Each line of the list corresponds to one directional power-on. Once the list is generated, it can be directly used in the next inspection cycle, allowing step 2 to reproduce the collection window and update the voxel-level moisture content map under the new main and auxiliary combination.

[0057] Setting the monitoring zone space and coordinates: A cuboid region with a longitudinal length of 6, a transverse width of 2, and a depth of 1 was selected on a section of EL-NSM hydrophobic modified soil subgrade. A rectangular coordinate system was established, with the longitudinal direction as x, the transverse direction as y, and the depth as z. Nodes and voxels: Coordinates 0, 2, 4, 6 were taken in the x-direction, 0, 2 in the y-direction, and 0.5, 1.5 in the z-direction, forming a node set of 16 nodes. The voxel grid consists of 3 intervals in the x-direction, 1 interval in the y-direction, and 1 interval in the z-direction, for a total of 3 voxels: V1 is located at x∈[0,2], y∈[0,2], z∈[0.5,1.5]; V2 is located at x∈[2,4], y∈[0,2], z∈[0.5,1.5]; V3 is located at x∈[4,6], y∈[0,2], z∈[0.5,1.5]. Sampling settings: Collect 10 equally spaced samples from each participating node within each acquisition window. Record the sample sequence number as follows: The sampling time step is 1.

[0058] Node Numbering and Coordinates: A three-segment LHD numbering system is used. Example: Node N(0,0,0.5) is numbered 0-0-05, node N(2,0,0.5) is numbered 2-0-05, node N(2,0,1.5) is numbered 2-0-15, and so on. Reference Resistance Diagram: All 16 nodes are sampled three times under dry conditions, and the mode is used to determine the reference resistance. The reference resistance values ​​for key nodes are given below, in ohms:

[0059] N(0,0,0.5): N(2,0,0.5): N(2,0,1.5): ;

[0060] N(2,2,0.5): N(4,0,0.5): N(6,0,0.5): Excitation grouping table: Generate vertical column groups along the vertical direction and horizontal row groups along the horizontal direction; use each column or row as the main excitation group entry, and use its adjacent columns or rows on both sides as secondary excitation group entries; the outermost ring is used as the guardrail group. Voxel neighbor node list: For the geometric center C1(1,1,1.0) of V1, take the vertical nearest node as N(0,0,0.5), the horizontal nearest node as N(2,0,0.5), and the deep nearest node as N(2,0,1.5). For the center C2(3,1,1.0) of V2, take the vertical nearest node as N(2,0,0.5), the horizontal nearest node as N(4,0,0.5), and the deep nearest node as N(2,0,1.5). For the center C3(5,1,1.0) of V3, take the vertical nearest node as N(4,0,0.5), the horizontal nearest node as N(6,0,0.5), and the deep nearest node as N(2,0,1.5).

[0061] set up The original resistance value of the i-th node at the k-th moment within a certain acquisition window; The reference resistance of the i-th node; The repositioning curve is defined as follows: , representing the amount of change relative to the reference; The curve after amplitude stretching is defined as:

[0062] ;

[0063] Linearly map the repositioning curve to 0 to 1; : The sign of the direction sequence, if If the judgment is superior, The judgment is a draw, if The judgment is as follows: : The first offset sequence number, which is the starting k that appears at least twice consecutively on the curve for the first time; Peak number, which is the first time the curve shows a transition from upward to downward (k). : Platform starting point number, which is the starting k where the curve first appears consecutively at least twice.

[0064] To make the calculation more intuitive, the original curves are constructed to meet the subsequent criteria, while the numerical variation is controlled within 1 ohm for easy addition to the reference resistance. The original curves of the V1 triple bond node in the vertical positive window are given below; representative nodes for other windows are given to reduce redundancy, and all are calculated using the same method. The closest vertical node of V1, N(0,0,0.5), is shown in the original curve of the vertical positive window:

[0065] ;

[0066] V1 has the nearest horizontal node N(2,0,0.5) and the original curve in the positive vertical window:

[0067] ;

[0068] V1's deepest nearest node N(2,0,1.5), the original curve of the vertical positive window:

[0069] ;

[0070] For N(0,0,0.5): ; Similarly, we can obtain the results for N(2,0,0.5) and N(2,0,1.5). , .

[0071] by For example, the direction sequence is obtained: According to the definition: ;right The first segment continues to rise, satisfying the condition that "the first segment of the direction sequence is upward"; for A slight rise followed by a rapid plateau is obtained, which is beneficial for subsequent comparison when passing over the guardrail. The same repositioning and stretching methods are used in the horizontal forward, vertical reverse, and horizontal reverse windows. To save space, the original values ​​and conclusions of the two representative curves V1 are given, and the verification method is the same:

[0072] A horizontal positive window, N(2,0,0.5) original curve The first segment of the directional sequence after repositioning and stretching is upward; the longitudinal reverse window, N(0,0,0.5) is the original curve. After returning to its original position and stretching, the price initially drops and then rises slightly, satisfying the termination criterion of "stabilizing after falling back".

[0073] For the four types of acquisition windows, sequentially set the window identifier, node number, and... , , The direction sequence summary is written into the time stack, denoted as... Where LF, XF, LR, and XR represent longitudinal forward, lateral forward, longitudinal reverse, and lateral reverse, respectively. Candidate conduction path templates are generated along the longitudinal direction. Its voxel sequence is Generate candidate conduction path templates along the horizontal direction. Its voxel sequence is a single segment , , (Because y has only one interval). Use in the vertical window. As a template set, it is used in landscape windows. Voxels are naturally mapped to their corresponding sequences based on shared surface relationships.

[0074] set up The i-th node's drop segment entries under a certain window satisfy either a decrease followed by stabilization or a decrease followed by increase. The segment entries of the i-th node under a certain window satisfy either an initial increase followed by stabilization or an initial increase followed by a decrease. , : These are the initial offset numbers for the dripping segment and the siphoning segment, respectively; The category of the response fingerprint pair of the i-th node within this window is defined as:

[0075] ;

[0076] in , This represents the peak index of the corresponding entry. The identifiable droplet segment begins at k=1 (at least two consecutive drops), and the scooping segment begins at k=3 (at least two consecutive lifts). ;get In the same window, the first segment of N(2,0,0.5) continues upwards, but the drop segment is not valid; only the siphon segment is valid, which is considered as no pairing. This node is used as the criterion for the first segment of the direction sequence. After a weak rise, the plateau in N(2,0,1.5) is only used as a reference for passing through the guardrail.

[0077] Under the vertical window, the vertical member N(0,0,0.5) of the triple bond group is of the forward-push type and its curve shape is of the upward-flat type; the first segment of the direction sequence of the lateral and deep members is upward, satisfying the axial penetration assignment rule in the claims. Therefore, V1 is assigned the water-containing behavior label as axial penetration under the vertical window, and the direction symbol is vertical, consistent with the window's energizing direction. Under the lateral window, V1 does not simultaneously satisfy the oblique condition of the forward-push type staggered peaks of the vertical and lateral dual windows, so it is not assigned oblique penetration. The same method is used to calculate V2 and V3. To give the complete link, its original curves are constructed so that V2 and V3 also satisfy the forward-push type of the vertical members and the upward first segment of the lateral and deep members under the vertical window, thus V1, V2, and V3 are... All of them are given axial penetration and the direction mark is longitudinal.

[0078] Using V1, V2, and V3 as seed voxels, the process proceeds according to shared surface relationships and longitudinal direction markings: V1 and V2 share an adjacency relationship of x=2 surfaces, and V2 and V3 share an adjacency relationship of x=4 surfaces. All three are merged into the same penetrating chain segment. Because the chain segment only connects in pairs and there is no three-chain convergence, no node volume is formed; the whole is considered a single penetrating volume. Among the voxels sharing surfaces with the penetrating volume, none are judged as interface seepage or bottom-to-bottom backflow voxels; therefore, the transition volume is an empty set; no connecting filling is required, and the filling volume is an empty set.

[0079] According to the mapping rules, the through-body corresponds to a high level, resulting in a voxel-level water content grade map: V1, V2, and V3 are all high levels.

[0080] This monitoring zone has a total of 3 voxels, with high-grade voxels accounting for a certain percentage. Furthermore, a continuous band-like structure with a length of 3 voxels is formed longitudinally, satisfying the criterion for high-level dominance. Therefore, the dominance level of this monitoring zone is set to high level. The output moisture level map of the monitoring zone is a single high-level block, with a continuous high-level band superimposed in the x-direction. Using a continuous segment without nodes as a chain segment, a single continuous chain segment is obtained. Its fields are: monitoring partition number is 1, starting voxel V1, ending voxel V3, direction symbol is vertical, chain segment length is 3, number of traversed voxels is 3, adjacent monitoring partition numbers are empty, and the number of acquisition window categories involved is 2 (both vertical forward and vertical reverse directions participate in the assignment). The main excitation group selection follows the principle of direction consistency priority: for The primary excitation group is selected based on the number of columns overlapping with the chain segment and containing the geometric center x=3. This column is located at the center between x=2 and x=4. The primary excitation group is selected from the excitation grouping table, choosing the column group closest to x=4. The secondary excitation group is selected from the column groups adjacent to it on both sides. The guardrail group is selected from the outer ring covering both ends of the through strip. The execution order follows a combination of risk-first and spatial adjacency: the through chain segment... The corresponding vertical forward window is placed in position 1, the vertical reverse window is placed in position 2, and the horizontal window, used for supplementary verification, is placed after it. The final excitation execution list for the next cycle contains 3 power-on instructions, namely vertical forward, vertical reverse, and horizontal forward. Each instruction records the sequence number, monitoring zone number, acquisition window direction, main excitation group entry, auxiliary excitation group entry, guardrail group entry, target penetration chain segment number, and target voxel range description.

[0081] Figure 2This paper demonstrates four typical resistance response curve shapes in this invention. These shapes are key technical features obtained through directional sequence analysis of the time series in the window curve book. The horizontal axis represents the acquisition time in seconds, with a test period of 160 seconds; the vertical axis represents the normalized resistance change, ranging from -1.0 to 3.5, obtained by repositioning and linearly stretching a reference resistance map. The flat upper curve (solid line) embodies the typical response characteristics of axial penetration, showing a continuous rise of at least three times starting from the first offset number (approximately 30 seconds), followed by a peak value at approximately 60 seconds and then remaining stable. The plateau starts at 90 seconds, after which a constant high resistance value is maintained. This curve shape corresponds to the longitudinal or lateral directional current-carrying effect generated by the main excitation group in the excitation grouping table. When the corresponding directional member in the three-bond group exhibits a forward-pushing or balanced response, it indicates significant axial penetration behavior in that voxel region. The up-and-down type curve (short dash) reflects the characteristic response pattern of interfacial seepage. It begins to rise at the same initial offset number, but shows a clear downward trend after the peak number, eventually returning to near the baseline. This curve segment with an initial increase followed by stabilization or an initial increase followed by a decrease is defined as the "drawing segment" entry. The paired "dropping segment" entry shows a pattern of initial decrease followed by stabilization or an initial decrease followed by an increase. Together, they constitute the back-drawing characteristic in the response fingerprint pair. The upper plateau type curve (medium dash) shows a composite response behavior of oblique penetration, characterized by two distinct rising phases separated by a brief plateau period. The first rise occurs between 30 and 60 seconds, followed by a plateau state maintained between 60 and 90 seconds. The second rise begins at 90 seconds and continues until the end of the test. This morphology corresponds to the forward-pushing response fingerprint pair, and the order of the peak numbers is inconsistent in the vertical and horizontal windows, indicating that water seeps in simultaneously in multiple directions. The upper-upper plateau type curve (long dash) represents the response characteristics of chamber water storage, showing a plateau state after at least four consecutive rises. The curve rises sharply in the first 60 seconds, then enters a long-term stable high-resistance plateau. When the response fingerprint pairs of the voxel under the four acquisition windows are all balanced, and at least two of the three-bond members are of type I morphology, the voxel is labeled as having chamber water storage behavior, corresponding to a relatively closed water accumulation space formed in the pore structure.

[0082] Figure 3This paper details the comparative characteristics of the resistance response of the voxel triple bond group under different excitation directions, verifying the effectiveness of the excitation grouping table and candidate conduction path template. The horizontal axis represents the acquisition time (test period 360 seconds), and the vertical axis represents the resistance offset in kiloohms, with a measurement range of -5kΩ to +11kΩ and a baseline of 0kΩ. The longitudinal excitation response curve (solid line) exhibits typical forward-pushing axial penetration characteristics, rising rapidly from the first offset sequence (approximately 60 seconds), reaching a maximum offset of +7kΩ at the peak phase (approximately 90 seconds), and then entering a plateau phase where it maintains a constant high resistance. This response pattern indicates that when the main excitation group in the excitation grouping table performs directional energization in a longitudinal sequence, the longitudinally nearest node in the triple bond group generates a significant resistance increment, corresponding to the body response confirmation of the longitudinal members in the voxel neighbor node list. Verification based on the guardrail pass markers shows that the amplitude range of the guardrail group curve is much smaller than half the amplitude range of the triple bond group curve, confirming the reliability of this response. The transverse excitation response curve (short dash) exhibits a backflow-type interface seepage characteristic. Its initial offset sequence is the same as that of the longitudinal excitation, but the peak sequence is significantly delayed, rapidly falling back to the baseline after reaching a peak of +5kΩ. This curve shape corresponds to the pairing pattern of droplet and pick-up segments, where the initial offset sequence of the droplet segment is earlier than that of the pick-up segment, and there is a time difference in the peak sequence, forming a typical backflow-type response fingerprint pair. This feature indicates that water seeps along the interface under transverse excitation, and the change in resistance from rising to falling reflects the redistribution process of pore water. The deep excitation response curve (medium dash) shows a bottom-backflow characteristic pattern. Initially, the resistance value drops to -4kΩ, then begins to rise and tends to stabilize after 150 seconds. This response corresponds to the leading effect of the peak sequence of the triple bond group under the deep window, where the morphology type of the deep member is type II, while the morphology types of the longitudinal and transverse members are type I, forming a unique bottom-backflow label. The oblique excitation response curve (dotted line) exhibits a balanced oblique penetration characteristic, maintaining a slow but continuous upward trend throughout the test cycle, eventually reaching a steady-state value of +6kΩ. This curve corresponds to a balanced response fingerprint pair, where the initial offset sequence and peak sequence of the droplet and drain segments remain consistent, indicating a synergistic response effect under multi-directional excitation. The guardrail group response curve (thin dotted line) serves as a reference, with its resistance offset consistently remaining within a small range of ±1kΩ, verifying the effectiveness of the guardrail group settings in the excitation grouping table and ensuring the accuracy and reliability of the measurement results.

[0083] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A method for monitoring the evolution of moisture content in hydrophobically modified soil subgrade during its service life, characterized in that, The method includes: Step 1: Install and number resistive soil moisture probes in the longitudinal, transverse and depth directions within the EL-NSM hydrophobic modified soil subgrade, and establish a node index table, geometric adjacency table and voxel grid; collect the reference resistance of each node under dry conditions and generate a reference resistance map; generate an excitation grouping table based on the geometric adjacency table, which includes a main excitation group, a secondary excitation group and a guardrail group, and generate a list of neighboring nodes for each voxel; Step 2: Perform directional energizing in four directions according to the excitation grouping table to generate four types of acquisition windows; align and linearly stretch the time series of each node using the reference resistance diagram, generating a directional sequence consisting of up, flat, and down curves, extracting and registering the three indices of the first offset, peak value, and platform start point; select the vertical nearest, horizontal nearest, and deep nearest nodes to form a three-bond group for each voxel based on the voxel's neighboring node list, and assemble the time stack under the four types of acquisition windows; generate candidate conduction path templates consistent with the acquisition window directions using the node index table and geometric adjacency table, and map the voxels to the template sequence according to the shared surface relationship; Step 3: Pair droplet entries with pickling entries within the same acquisition window; within the same acquisition window, define the curve segments that show an initial increase followed by stabilization or an initial increase followed by a decrease relative to the main excitation polarity as pickling entries, and define the curve segments that show an initial decrease followed by stabilization or an initial decrease followed by an increase relative to the main excitation polarity as droplet entries, and pair them accordingly; based on the fingerprint category and the curve morphology of the three-bond group members, assign water-containing behavior labels to voxels on the candidate conduction paths and generate direction marks; perform riveting and penetration on voxels of the same type along the direction marks using the adjacent relationship of the shared surface to form a through body and a transition body, and finally obtain a voxel-level water content level map; Step 4: Summarize the voxel-level moisture content maps into a monitoring zone moisture content map and a list of connected segments. Based on the spatial orientation of the connected segments, select the main incentive group, auxiliary incentive group, and guardrail group combination from the incentive grouping table for the next inspection, generate an incentive execution list, and repeat the execution in subsequent inspection cycles to continuously output the moisture content evolution results during the service period.

2. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 1, characterized in that, Step 1 specifically includes: Installing and numbering resistive soil moisture probes longitudinally, laterally, and in depth within the EL-NSM hydrophobic modified soil subgrade; establishing a node index table, which records the longitudinal, lateral, and depth coordinates of each numbered node; generating a geometric adjacency table based on the grid relationships of nodes in the three directions, listing the adjacent node numbers that share edges or faces with each node; dividing the monitoring volume into a voxel grid, and selecting one nearest node in each of the three directions for each voxel to form a list of adjacent nodes; collecting the resistance of each node under dry conditions to form a reference resistance map; generating an excitation grouping table based on the row and column distribution of the node index table, which divides nodes into three categories: main excitation group, auxiliary excitation group, and guardrail group, and listing the energizing sequence in both longitudinal and lateral sequences.

3. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 2, characterized in that, In step 2, based on the excitation grouping table, two sets of window numbering tables, one for the longitudinal sequence and one for the transverse sequence, are generated within the monitoring zone. Directional energizing excitation is performed sequentially in the four directions: longitudinal forward, transverse forward, longitudinal reverse, and transverse reverse, forming four types of acquisition windows. For each acquisition window, the numbers of the main excitation group nodes, auxiliary excitation group nodes, and guardrail group nodes involved in the window are recorded according to the window numbering table. The resistance time series of the above nodes is collected to form the original curve book.

4. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 3, characterized in that, In step 2, for each time series in the original curve book, the reference resistance of the corresponding node in the reference resistance diagram is found, and the reference resistance is subtracted from the time series point by point to obtain the repositioning curve. Within each acquisition window, the minimum value of the homing curve is taken as the zero point and the maximum value as one unit. The homing curve is linearly stretched to obtain the amplitude stretching curve and recorded in the window curve book. This step directly uses the reference resistance diagram to complete the homing reference, and the result is written into the window curve book.

5. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 4, characterized in that, In step 2, the direction sequence generation process includes: for each curve in the window curve book, generating a direction sequence point by point according to the size relationship of adjacent sampling points: the later sample is greater than the previous sample and is marked as up, the later sample is equal to the previous sample and is marked as flat, and the later sample is less than the previous sample and is marked as down; based on this, three position indicators are extracted and registered in the indicator table: the first offset sequence number is the position where the first consecutive up occurs at least twice, the peak sequence number is the turning point where the first up occurs and then the next down occurs, and the platform start sequence number is the position where the first consecutive flat occurs at least twice; based on the direction sequence and the three position indicators, the curve is determined to be one of the following four types and registered in the type table: up-flat type, characterized by at least three consecutive ups followed by at least two consecutive flats; up-down type, characterized by at least three consecutive ups followed by at least two consecutive downs; up-platform type, characterized by at least two consecutive ups, followed by at least two consecutive flats, followed by at least two more consecutive ups; up-up-platform type, characterized by at least four consecutive ups, followed by at least one flat.

6. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 5, characterized in that, In step 2, each voxel in the voxel mesh is traversed, the list of neighboring voxels is called, and the triple bond group of the voxel is extracted. The triple bond group consists of the vertical nearest node, the horizontal nearest node, and the deep nearest node. Under the four types of acquisition windows, the entries of the triple bond group members in the index table and the morphology table are read respectively, assembled into a time stack according to the window order, and written to the voxel workbook. This step directly uses the voxel mesh and the list of neighboring voxels to complete the binding of voxel and node features.

7. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 6, characterized in that, In step 2, based on the coordinate relationships in the node index table and the adjacency pairs in the geometric adjacency table, three types of candidate pathway templates are generated along the longitudinal, lateral, and depth directions, respectively. Each template consists of a sequence of voxels of a shared surface. The voxel sequences are inherited as the adjacency relationships of the shared surface at the voxel level through the geometric adjacency table. For each acquisition window, a path template with the same direction as the window is selected, and the voxel sequences in the path template are registered as the candidate pathway set for that window. This step explicitly uses the node index table to generate the direction and uses the geometric adjacency table to implement the adjacency relationships at the voxel level.

8. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 7, characterized in that, In step 2, within the voxel workbook, drop and draw segments for each voxel in the same acquisition window are paired to form response fingerprint pairs. Each drop and draw segment entry consists of an initial offset number, a peak number, and a morphological type. The response fingerprint pair category is determined according to the following rules and written into the fingerprint matrix: Forward type: the initial offset number of the drop segment is earlier than that of the draw segment; Backward type: the initial offset numbers of the drop and draw segments are the same, and the peak number of the drop segment is later than that of the draw segment; Balanced type: the initial offset numbers and peak numbers of the drop and draw segments are identical. Simultaneously, referring to the guardrail groups in the excitation grouping table, the amplitude range of the guardrail group curve and the amplitude range of the triple bond group curve are calculated within the same window. If the amplitude range of the guardrail group is less than half of the amplitude range of the triple bond group, then the voxel is registered with a guardrail pass mark for subsequent ontology response confirmation in label assignment. This step explicitly uses the contents of the voxel workbook, the excitation grouping table, and the window curve book.

9. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 8, characterized in that, In step 3, within each acquisition window, for voxels in the candidate conduction path set, their category in the fingerprint matrix and type in the morphology table are read. Combined with the guardrail passmark, water-containing behavior labels and direction marks are generated at the voxel level according to the following rules and written to the label layer: Axial penetration: When the window direction is vertical and the vertical member in the triple bond group is either forward-pushing or balanced, and the morphology type of the vertical member is type I or IV, while the first segment of the direction sequence of the lateral and depth members is upward, then the voxel is marked as axial penetration, and the direction mark is vertical and consistent with the window direction; when the window direction is horizontal and the lateral member in the triple bond group is either forward-pushing or balanced, and the morphology type of the lateral member is type I or IV, while the first segment of the direction sequence of the vertical and depth members is upward, then the voxel is marked as axial penetration, and the direction mark is vertical and consistent with the window direction; when the window direction is horizontal and the lateral member in the triple bond group is either forward-pushing or balanced, and the morphology type of the lateral member is type I or IV, while the first segment of the direction sequence of the vertical and depth members is upward, then the voxel is marked as axial penetration, and the direction mark is... Horizontal and aligned with the window direction; Diagonal penetration: When the voxel is in a forward-pushing type under both the vertical and horizontal windows, and the order of the peak numbers in the two windows is inconsistent, the voxel is marked as diagonal penetration, and the direction symbol is a combination of vertical and horizontal; Interface seepage: When the deep member of the triple bond group under the deep window is in a backflow type and the morphology type is type II, the voxel is marked as interface seepage, and the direction symbol is deep; Bottom backflow: When the peak number order of the triple bond group under the deep window is led by the deep member, and the morphology type of both the vertical and horizontal members is type I, the voxel is marked as bottom backflow, and the direction symbol is deep; Chamber water storage: When the response fingerprint pairs of the voxel are all balanced under all four types of windows, and at least two of the triple bond group members are type I, the voxel is marked as chamber water storage, and the direction symbol is omnidirectional.

10. The method for monitoring the evolution of moisture content in hydrophobic modified soil subgrade during its service life as described in claim 9, characterized in that, In step 3, voxels marked as axial or oblique penetration in the label layer are used as seed voxels. Based on the geometric adjacency list, penetration chains are generated at the voxel level using shared face adjacency relationships, pointing along their respective direction marks. When adjacent voxels at the front end of a penetration chain segment have the same direction mark and are marked as axial or oblique penetration in the label layer, that voxel is merged into the chain segment. When three or more penetration chains meet on the same voxel and their direction marks are different pairwise, that voxel is registered as a node and a riveting operation is performed. The riveting operation merges the chains. The end of the segment is bound to the node body, forming a component of a single through-body; subsequently, voxels that share the surface with the through-body and are labeled as interface seepage or bottom backflow are incorporated into the transition body; the remaining voxels that are in contact with the through-body or transition body through the shared surface are subjected to joint filling to form the fill body; the through-body, transition body and fill body are mapped to a voxel-level moisture content grade map: the through-body corresponds to the high grade, the transition body corresponds to the medium grade, and the fill body corresponds to the low grade; this step explicitly uses a geometric adjacency table and voxel mesh to complete spatial connection and solidification.

Images